You are viewing a plain text version of this content. The canonical link for it is here.
Posted to openrelevance-dev@lucene.apache.org by "Simon Willnauer (JIRA)" <ji...@apache.org> on 2009/11/13 12:10:39 UTC

[jira] Commented: (ORP-1) Use existing collections for relevance testing

    [ https://issues.apache.org/jira/browse/ORP-1?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12777453#action_12777453 ] 

Simon Willnauer commented on ORP-1:
-----------------------------------

Good stuff robert!  I guess we should split it up and have one issue for the basic stuff like base ant scripts and LICENCE files etc. and another one for the first collection code.

thoughts?

> Use existing collections for relevance testing
> ----------------------------------------------
>
>                 Key: ORP-1
>                 URL: https://issues.apache.org/jira/browse/ORP-1
>             Project: Open Relevance Project 
>          Issue Type: New Feature
>          Components: Collections, Judgments, Queries
>            Reporter: Robert Muir
>            Assignee: Simon Willnauer
>         Attachments: ORP-1.patch
>
>
> I created a list of existing collections with queries and judgements on the wiki here: http://cwiki.apache.org/ORP/existingcollections.html
> These can be downloaded from the internet.
> (please add more if you know)
> I've created source code (ant and java) to download these collections, and reformat them to the trec format that the lucene benchmark expects.
> each collection has its own ant script to download the collection, and java code to reformat, although I have some shared code at the top level.
> The resulting output for each collection is a "corpus.gz" file, a queries file, and a judgements file, 
> The corpus.gz is a gzipped file that can be indexed with ant via the lucene benchmark package (using TrecContentSource)
> Once the index is created, the command-line tool QueryDriver under the lucene benchmark quality/trec package can be used to run the evaluation.
> It will print some summary output to stdout, but will also create a submission file that can be fed to trec_eval.
> For starters, I will only have support for one collection in the patch, the Indonesian "Tempo" collection (around 23,000 docs)
> We can simply add subdirectories for additional collections (it does a contrib-crawl like thing).
> Once I finish wrapping up some documentation (such as description of the formats, some javadocs, and an example), I'll upload the patch.
> These formats are actually documented in the lucene-java benchmark package already, but I think it would be nice to add this for non-java users.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.